Best practices for right-sizing Amazon OpenSearch Service domains

This article provides comprehensive guidance on right-sizing Amazon OpenSearch Service domains to balance performance, scalability, and cost-efficiency.

• Monitor key CloudWatch metrics: CPU utilization (60-80%), JVM memory pressure (65-85%), storage utilization (70-85%)
• Under-provisioning signs: sustained high CPU/memory, increasing latency, search timeouts, circuit breaker exceptions
• Over-provisioning signs: low CPU/memory utilization (<40-50%), excessive free storage (>70%)
• Remediation strategies: delete unused indexes, scale vertically/horizontally, optimize shard counts (10-50 GB each)
• Use ISM for automated cleanup, force merge for read-only indexes, UltraWarm/cold storage for older data
• Establish 2-4 week baselines, conduct weekly reviews initially, then monthly for stable workloads
• Optimize shard configuration: keep <20 shards per GB JVM heap, maintain even distribution across nodes
• Deploy CloudWatch alarms for proactive monitoring and early stress detection
• Right-sizing is continuous; regularly review metrics and adjust as workloads evolve

Right-sizing OpenSearch domains requires ongoing monitoring and optimization using CloudWatch metrics, shard analysis, and scaling strategies to maintain performance without over-provisioning.